Basic ENS Products
Basic products display only the raw ENS forecast data, without any particular modification or post-processing. Individually plotted forecast charts of MSLP and 850hPa temperature of all the ensemble members, including the HRES and the Control, can be displayed individually, but for ease of visual comparison are displayed together as "Postage Stamps” (or "Postage Stamps Maps or Charts"). They cover a limited area of the globe, normally Europe and eastern North Atlantic. The charts are intended to be used for reference - for example to explain the spread in synoptic terms and in particular, the reasons for extreme weather etc. It is not possible to select the ENS member with the best subsequent forecast since the performance of any member during the first 12hrs of the forecast has little relevance to its skill beyond T+48hrs in the same area.
In order to compress the amount of information being produced by the ENS and highlight the most predictable parts, individual ENS members that are "similar" according to some measure (norm) can be grouped together. This process is known as Clustering.
Fig220.127.116.11: A example of postage stamps showing the PMSL forecasts from HRES, ENS CTRL, and all 50 ENS members, data time 00UTC 19 May 2017, verifying time T+120hr at 00UTC 24 May 2017. Some large differences in the pressure pattern can be seen on individual ENS members. Each member has been allocated a cluster, shown in a different colour above each frame for lead-times T+120 and above only. The representative member of each cluster is shown by arrows. The clusters are shown in Fig18.104.22.168.
Fig22.214.171.124: Clustering for the case shown in Fig126.96.36.199. The three clusters for T+120hr are in the left column. Clustering is based on 500hPa geopotential height pattern.
Spaghetti diagrams are available on ecCharts and display isolines for Surface Pressure or 500hPa geopotential height or 850hPa temperature, the value for which is chosen by the user. The isolines, drawn for each member, are initially very tightly packed, but they increasingly spread out as the forecasts progress, reflecting the flow-dependent increase in forecast uncertainty. Spaghetti Diagrams are sensitive to gradients; in areas of weak gradient they can show a large spread of the isolines, even if the situation is highly predictable. Conversely, in areas of strong gradient they can display a small spread of the isolines, even if there are important forecast variations.
Fig188.8.131.52: Spaghetti Plot 500hPa geopotential height plotted at 560dam for T+30hr. ENS data time 12Z 26 May 2017. The ENS members (grey) are quite tightly packed except at about 20°W where gradients are slack and there is some uncertainty in the trough disruption. The control member is shown in Red and HRES in Blue.
Fig184.108.40.206: Spaghetti Plot 500hPa geopotential height plotted at 560dam for T+144hr. ENS data time 12Z 26 May 2017. The ENS members have become more spread out, but retain the general pattern of an upper ridge over the northwest Atlantic and another over the North Sea but there are differences in position (or speed of movement eastwards) and amplitude. The control member is shown in Red and HRES in Blue.
Ensemble Mean and Median
The ensemble mean (EM) forecast is a simple but effective product as the averaging serves as a filter to reduce or remove atmospheric features that differ amongst the members. The EM tends to weaken gradients; all members might forecast an intense low-pressure system with gale force winds but in different positions, but the EM will only show a rather shallow spread out depression giving the impression of weak average winds. High-impact events which in the EM appear weak or absent can be easily overlooked or at best regarded as less predictable. The EM may show only a weak feature but it does not mean that there is no noteworthy development in any member, nor that the more developmental outcomes are themselves less probable, since all evolutions shown by ENS members are equally likely. Inspection of the Postage Stamps and/or use of probabilities is therefore essential in conjunction with examining the EM.
The EM is most suited to parameters like temperature and pressure, which usually have a rather symmetric Gaussian distribution at each point. In the short range the EM is very similar to the CTRL or HRES due to the anti-symmetry (equal positive and negative) of the initial perturbations. The EM is less suitable for wind speeds and precipitation because these exhibit skewed distributions and for these parameters the ensemble median might be more useful. The ensemble median is defined as the value of the middle ensemble member where the members have been ranked by value.
Fig220.127.116.11: Ensemble Mean PMSL (Red) and Spagetti Plot of 990hPa isobars (Grey) from ENS data time 00UTC 10 June 2017 T+120 forecast verifying at 00UTC 15 June 2017. There is a wide diversity amongst ENS members in the location and shape of the depressions. The ensemble mean depression is smooth by comparison and less deep (the inner isobar has a value of 995hPa). Because of averaging of the ENS members, the pressure gradients and associated winds will generally be less strong in the ensemble mean field than in the ENS members themselves. Chart taken from ecCharts.
Fig18.104.22.168: As Fig22.214.171.124 but with the spread of PMSL by ensemble members (coloured - orange: high spread, blue: low spread). Highest spread of PMSL is on the eastern side of the depression indicating greater uncertainty in the strength of a southerly wind. There is lower spread to the west where most members suggest a fairly low pressure and higher probability of a northerly flow. The smallest spread is near the centre of the depression indicated by the ensemble mean but wind direction is very uncertain here; it depends upon the positioning of the low in the of individual ENS members. Chart taken from ecCharts.
The most consistent way to convey forecast uncertainty information is by the probability of the occurrence of an event. The event can be general or user-specific regarding exceeding an event threshold. The event threshold may corresponds to the point at which the user has to take some action to mitigate potential damage from a significant weather event. Probabilities can be instantaneous (e.g. 10m wind probabilities), or be calculated over a time interval (e.g. precipitation) because the values are themselves originally computed as values accumulated over some time interval. Probabilities for extreme wind gusts are computed as probabilities over 24 hours because it is considered more important to know that an extreme wind gust might occur than to know the precise time.
Fig126.96.36.199: As Fig188.8.131.52 with the probability of wind ≥10m/s. There is a higher probability (dark blue) in the area west of Ireland where the pressure gradient is uncertain although the direction is fairly certain. The ENS mean would not suggest such strong winds. There is very low probability (white) south of Greenland where the gradient is generally light and the direction is uncertain. Chart taken from ecCharts. Light blue >25%, Blue >50%, Dark blue >75% probability.
ecCharts allows probability of a combination of events occurring together to be displayed, such as strong wind gusts and heavy snowfall (for assessment of drifting snow), precipitation and surface temperature (to aid rain/snow forecasting), strong winds and significant wave height (to help forecast dangerous conditions at sea).
Fig184.108.40.206: As Fig220.127.116.11 with the probability of significant wave height ≥4m. Chart taken from ecCharts. Yellow >25%, Orange >50% probability.
Fig18.104.22.168: As Fig22.214.171.124 with the probability of significant wave height ≥4m AND wind ≥10m/s. Chart taken from ecCharts. Light blue >10%, Green >20%, Yellow >30% probability.
Forecast expressed in terms of intervals
Forecast intervals (e.g. “temperatures between 2°C and 5°C”, or “precipitation between 5 and 8mm/24hr”) can be used as a hybrid between categorical and probabilistic forecasts. ecCharts provide a simple way of displaying probabilities above or below thresholds and by intercomparison can give a indication of probability of a parameter lying between the thresholds. For example for maximum temperatures at Vilnius, from Fig126.96.36.199 there is a 20% probability of being ≥20°C and from Fig188.8.131.52 there is a 25% probability of being ≤15°C. Therefore there is a 55% probability that the maximum temperature will lie between these two values. Other combinations of parameters are possible (e.g.The probability of combined events of wind gust and total snowfall is available on ecCharts as an aid to forecasting drifting of snow).
Fig184.108.40.206: Chart taken from ecCharts showing ENS probability of maximum 2m temperature ≥20°C (ecCharts colour bands for this scheme denote 5-20-40-60-80-95-100%). There is a 20% probability of maximum temperatures ≥20°C at Vilnius (shown in the box). The location of Vilnius is shown by the pin.
Fig220.127.116.11: Chart taken from ecCharts showing ENS probability of maximum 2m temperature ≤15°C (ecCharts colour bands for this scheme denote 5-20-40-60-80-95-100%). There is a 25% probability of maximum temperatures ≤15°C at Vilnius (shown in the box). The location of Vilnius is shown by the pin.
Probability of Combined Events
Several charts on ecCharts are available to show the probability of combined events with thresholds under user control. The probability is computed as the ratio of the number of the ensemble members in which both event conditions are met to the total number of ensemble members. The current available charts are probabilities of:
- 2m temperature and total precipitation,
- wind speed and total precipitation,
- wind gust and total snowfall,
- 10m wind speed and significant wave height.
Fig18.104.22.168: Chart taken from ecCharts showing ENS probability of wind gust ≥10m/s and ≥2mm/12hr (ecCharts colour bands for this scheme denote Light blue 5-35%, Blue 35-65%, Dark blue 65-95%, Purple >95%). There is a 31% probability of exceeding the thresholds at Munich (shown in the box). The location of Munich is shown by the pin.
Additional Sources of Information
(Note: In older material there may be references to issues that have subsequently been addressed)
- Watch a detailed theoretical presentation regarding Statistical Post-Processing of Ensemble Forecasts (30sec delay before start).